标签：#embeddedsystems

Removing expf() from a fire detector: one header, 1.95x faster, zero accuracy loss A smoke detector is not a demo project. When it fires, someone either evacuates in time or doesn't. The firmware running on that microcontroller has one job, and it needs to do it without hesitation, without bloat, and without dependencies that can fail in unexpected ways. Last May 28th I published a bare-metal fire detection system built with Hasaki 刃先 — a neural network trainer that exports standalone C headers with no runtime, no Python, no TensorFlow. The model is a 12-8-4-1 MLP trained on 28,596 sensor readings. It fits in 3.8 kB of Flash and achieves 99.93% accuracy on held-out data, with a single missed fire event out of 3,599. But there was something in that header that bothered me. static inline float sigmoid ( float x ) { return 1 . 0 f / ( 1 . 0 f + expf ( - x )); } expf() . Right there in a life-safety application. On a microcontroller that may not have a hardware FPU. The problem with expf() on bare metal On processors with a hardware FPU — like the ESP32-C3 — expf() is fast. But the moment you deploy to an ATmega328P, an ATtiny85, or any Cortex-M0 target, that call becomes software floating-point. The CPU has to simulate the operation in firmware, cycle by cycle. It works. But it carries hidden cost: unpredictable latency, dependency on math.h , and a transcendental function sitting in the critical path of every single inference. For a smoke detector running at 1 Hz this might seem irrelevant. But inference latency compounds with sensor reads, normalization, and communication overhead. And more importantly — if you're deploying to a truly constrained target, expf() might be the difference between fitting in Flash or not. The fix: one header from kigu-quant kigu-quant(comming soon) is a new tool in the Rosito Bench ecosystem. It generates ready-to-include C headers for evaluating mathematical functions on microcontrollers — no FPU, no libm, no dependencies. One command: k

The chatbot is tricking me!!! 💬📜⌛ When you text a chatbot, it doesn’t actually remember who you are or what you said two minutes ago. The exact millisecond it finishes typing a response, its brain completely wipes clean. To pull off the illusion of a continuous, flowing conversation, the web application secretly copy-pastes the entire past chat history, bundles it up, and blasts that whole massive block of text back into the processor every single time you hit send. Your "chat session" is an illusion maintained entirely by an ever-growing stateless prompt wrapper. You aren't interacting with a growing, adapting mind; you are repeatedly gas-lighting a brand-new entity into believing it has been talking to you for an hour. Wait, I am the one training it ??? 🚦🚸🚲 AI models are inherently blind to context; a computer doesn't instinctively know that a specific cluster of raw pixel values represents a real-world object. It requires billions of examples to be manually labeled by a human mind before the math can understand it. Every time you click on squares containing "traffic lights," "crosswalks," or "bicycles" to unlock a website, you are acting as an unpaid data annotator. You are manually labeling complex, messy real-world data points that feed directly into the computer vision systems of autonomous vehicles. The grand paradox of modern cyber security is that we force humans to act like mechanical data annotators to prove they are not computers, all so that computers can learn how to perfectly impersonate humans. The supercomputer is stupider than a toddler... 🍓👶🏻🖥️ We assume AI read letters and words the same way human eyes scan a page. It doesn't—it is entirely alphabet-blind. Before text hits the AI's brain, a parser chops strings of text into numerical blocks called "tokens." For example, the word "strawberry" isn't seen by the model as ten distinct letters; it is compressed into numerical IDs representing chunked pieces like "straw" and "berry". Because it never s